Small Target Detection Based On Background Reconstruction And Non-equilibrium Graph Cut In Infrared Image

Infrared small target detection technology is one of the key technologies of infraredimaging system, and the performance of the algorithm plays a key role on the operatingdistance and intelligent degree of infrared imaging guidance. Therefore infrared smalltarget detection technology in the military has important significance and value.The target characteristics and background characteristics in the infrared images havea significant difference, and this is the basic condition of background suppressiontechnique. This paper introduces a new type of top-hat transform technique, and thedifference between it and classic top-hat transform technique is that, the new top-hattransform technique takes two different but closely related structure elements to dooperations, and make full use of the difference between the target area and its surroundingbackground region, so it can inhibit the background better. The key of the top-hattransformation technology is the selection of structure element, and the different structureelements will get different effects, so the shortcoming of this kind method need betterprior knowledge, and if the SNR of the image is very low, only a single frame by thetop-hat transformation technology is difficult to detect the target.This paper proposes a new background suppression technique, and it can beautomatic achieve without any prior knowledge constraint advantages. And it is from theperspective of pixel neighborhood. Firstly, segment any two adjacent framesf₁andf₂to many blocks, and to one pixel block fromf₁, search its nearest neighbor blocks fromits adjoining framef₂, and then use these nearest pixel blocks to reconstruct this pixelblock and meet the minimum reconstruction error. Do the same processing to every blockoff₁, and finally get the background image reconstruction. At this time, the backgroundcan be reconstructed very well, but the reconstruction error of the target is bigger, so wecan remove most of the interference of background, making the target emerges. If thebackground ups and downs is not very intense, and the infrared target is small, this methodwill get good results.This paper also proposes a new image segmentation method——Minimum non-equilibrium graph cutting algorithm. It from the point of view of cutting energy,search for the optimal thresholds, and can segment the target more accurately.For the stable false target interferences in sequence images, this paper adoptsadaboost learning way to identify the real target and the false targets, and realize theinfrared small target low false alarm rate detection. This kind of method can get goodrefinement results. But this method has bad real-time.